Integrated Ad Delivery Planning for Targeted Display Advertising
Operations Research and Operations Management
Huaxiao Shen, Yanzhi David Li, Youhua Frank Chen, Kai Pan
Published in Operations Research, June 2020
Online publishers such as Yahoo or Facebook sells their advertising resources via either an upfront market or a spot market. (In the upfront market, the publisher enters into a contract with each advertiser, guaranteeing an agreed-upon number of impressions over a specified period at a fixed price. These contracts often vary in the length of the period set for the ads; therefore, at any given time, the publisher has many contracts, each of which has a varying time span. In the spot market, the publisher auctions off its remaining advertising resources in real time among advertisers who, either purposefully or spontaneously, want to display their ads during the current period.) The current practice of ad delivery suffers from two major inefficiencies. Firstly, separate management of the upfront market and spot market ignores the value of coordination and integration and thus misses potential profit. The second inefficiency is related to the selling approach in the spot market. Although theoretically appealing, simple auction mechanisms (e.g., the generalized second-price auction) have been found to have many drawbacks in practice.
Professor Yanzhi David Li, and Professor Youhua Frank Chen of City University of Hong Kong and co-authors Huaxiao Shen and Kai Pan propose integrating ad delivery planning for guaranteed ads and spot market ads. They study the operational decisions of an online advertising publisher, namely, how the publisher should allocate its advertising resources to different guaranteed ads and to spot market ads, considering the uncertain supply of ad resources, the requirement of guaranteed ads, and the bids submitted from advertisers in the spot market.
“The proposed model incorporates various practical concerns as constraints and allows the publisher to balance the trade-off between short-term profit and long-term benefit,” they say.
The authors derive a closed-form expression for the expected reach and then construct a distributionally robust chance-constrained programming model that maximizes the spot-market revenue subject to the attainment of contracted delivery and higher reach (measured by the unique audience an ad campaign is exposed to). The solution to their model is an ad delivery plan that is robust to the uncertainties of advertising resources. They propose a robust optimization model to safely approximate the original model, which is further transformed into a linear program for tractability. They develop a clustering algorithm to efficiently solve large-scale instances of the linear program.
In practice, the problem is often of a large scale and is challenging to solve. Computationally efficient algorithms are highly demanded for real implementations of their framework. They test the approach on random test cases and two real data sets. An efficient ad serving procedure is designed to incorporate additional constraints such as frequency caps and exclusivity. They show that their approximation of chance constraints is very tight. They explore the trade-off between short-term spot market revenue and the effectiveness of guaranteed ads through numerical experiments. The proposed integrated planning of guaranteed and non-guaranteed ads is shown to be capable of achieving higher profit and also ensuring effective delivery of guaranteed ads.
